[U-04] Geospatial Applications for Societal Benefits

[U04-02] Modelling Carbon Sequestration of Grass at Laboratory Utilizing Spectral Data

*San Lin Phyo¹、Yukihiro Takahashi¹、Ye Min Htay¹ (1.Hokkaido University)

キーワード：Carbon Sequestration, Remote Sensing, Normalized Difference Spectral Index (NDSI), Spectrometer, Machine Learning

Global climate changes are becoming worse as increasing of atmospheric concentration of greenhouse gases such as carbon dioxide (CO2), chlorofluorocarbons (CFCs), methane (CH4), and nitrous oxide (N2O). The contribution of each gas to the greenhouse effects is CO2 - 55%, CFCs - 24%, CH4 - 15%, and N2O - 6% [Demirbas, 2008]. [Demirbas, 2005] estimated that world carbon dioxide emission can be over 10000 million metric tons in 2025. In nature, forests, wetlands, grass, and croplands uptake carbon dioxide from the atmosphere in photosynthesis and store it for a long time. Wetlands soil carbon storage percentage is about 20% higher than forests ,65% than grass/shrub-lands [Liu et al., 2021]. Paddy fields are temporary wetlands and contain substantial CH4 emissions, contributing about 10% of all anthropogenic CH4 emissions [Nazaries et al., 2013]. However, [Liu et al., 2021] mentioned that the carbon storage of paddy soil is over 20% greater than other staple crops and grassland. Carbon storage of rice fields is changing at each growing stage. Although several research regarding carbon fixation of rice fields utilizing remote sensing have been published, there are still limitations regarding band selection to improve estimation of carbon sequestration. Therefore, this work focuses new ways to select the best appropriate bands for carbon fixation prediction. The objective of this work is to establish a model that can predict carbon sequestration using spectral data. In this experiment, the oat grass (pet eat grass) instead of rice plant was utilized because oat grass is same species of paddy plants and the characteristics of plants are also near the same. At first, the reflectance of grass was measured at laboratory using spectrometer and then calculated Normalized Difference Spectral Index (NDSI). Based on spectral indices, the models for carbon fixation were established applying machine learning algorithm. After that, carbon sequestration of grass was predicted using the models created and finally, the accuracy between predicted carbon sequestration and actual data was examined. In this meeting, the methods that were used for this experiment will be presented and the analysis results will be discussed.